= TGV divergence splitting
{c}
For a quadratic data term and divergence maps $A_1,A_2$, the dual constraint $A_1v\in D$ can be split through $\delta_D(A_1v)=\sup_w[\langle w,A_1v\rangle-\delta_D^*(w)]$. For a product of Euclidean row balls of radius $\beta$, its <support function> is $\delta_D^*(w)=\beta\sum_i\|w_i\|_2$. The resulting saddle coupling is $K(u,w)=A_2^*u-A_1^*w$. The <Chambolle–Pock algorithm> then uses a row-ball projection, a quadratic <proximal operator> and <radial soft thresholding>, with no projection onto an intersection involving a divergence operator.
Back to article page
